Mani Srivastava UCLA - EE Department Room: 6731-H Boelter Hall Email: mbs@ee.ucla.edu Tel: 310-267-2098 WWW: mbs Copyright 2004.
Post on 20-Dec-2015
217 Views
Preview:
Transcript
Mani SrivastavaUCLA - EE DepartmentRoom: 6731-H Boelter HallEmail: mbs@ee.ucla.eduTel: 310-267-2098WWW: http://www.ee.ucla.edu/~mbs
Copyright 2004 Mani Srivastava
Power-aware Design - Part IEnergy Consumers & Sources
EE202A (Fall 2004): Lecture #8
2 Copyright 2004 Mani Srivastava
Reading List for This Lecture Required
D. Rakhmatov, S. Vrudhula, and D. Wallach, “A model for battery lifetime analysis for organizing applications on a pocket computer”, IEEE Transactions on VLSI, to appear.http://nesl.ee.ucla.edu/pw/ee202a/rakhmatov-tvlsi-2003.pdf
Thomas Martin and Daniel Seiwiorek, "Non-Ideal Battery Behavior and Its Impact on Power Performance Trade-offs in Wearable Computing," Proceedings of the 1999 International Symposium on Wearable Computers, San Francisco, CA, October 18-19, 1999; pp. 101-106.
http://www.ece.vt.edu/~tlmartin/papers/nonideal_formatted.pdf
Recommended V. Raghunathan, C. Schurgers, S. Park, and M. Srivastava, "Energy-aware
Wireless Microsensor Networks," IEEE Signal Processing Magazine, March 2002. p. 40-50.
Others None
3 Copyright 2004 Mani Srivastava
Why worry about power?Intel vs. Duracell
No Moore’s Law in batteries: 2-3%/year growth
Processor (MIPS)
Hard Disk (capacity)
Memory (capacity)
Battery (energy stored)
0 1 2 3 4 5 6
16x
14x
12x
10x
8x
6x
4x
2x1x
Improvement(compared to year 0)
Time (years)
4 Copyright 2004 Mani Srivastava
Current Battery Technology is Inadequate
Example: 20-watt battery NiCd weighs 0.5 kg, lasts 1 hr, and costs $20 Comparable Li-Ion lasts 3 hrs, but costs > 4x more
Battery Rechargeable? Wh/lb Wh/litreAlkaline MnO2 NO 65.8 347Silver Oxide NO 60 500Li/MnO2 NO 105 550Zinc Air NO 140 1150NiCd YES 23 125Li-Polymer YES 65-90 300-415
5 Copyright 2004 Mani Srivastava
The Showstopper: Energy Need long lifetime with battery operation
No infrastructure, high deployment & replenishment costs Continual improvement in functionality, size, weight, and power
1.6x/year in DSP power sensing and RF components based on MEMs
But energy to wirelessly transport bits is ~constant
Shannon, Maxwell fundamental limit on ADC speed*resolution/power no Moore’s law for battery technology
~ 5%/year
QuickTime™ and aTIFF (Uncompressed) decompressorare needed to see this picture.
QuickTime™ and aTIFF (Uncompressed) decompressorare needed to see this picture.
Single-chip WirelessSensor Node
The Future
6 Copyright 2004 Mani Srivastava
Comparison of Energy Sources
Power (Energy) Density Source of Estimates
Batteries (Zinc-Air) 1050 -1560 mWh/cm3 (1.4 V) Published data from manufacturers
Batteries(Lithium ion) 300 mWh/cm3 (3 - 4 V) Published data from manufacturers
Solar (Outdoors)
15 mW/cm2 - direct sun
0.15mW/cm2 - cloudy day. Published data and testing.
Solar (Indoor)
.006 mW/cm2 - my desk
0.57 mW/cm2 - 12 in. under a 60W bulb Testing
Vibrations 0.001 - 0.1 mW/cm3 Simulations and Testing
Acoustic Noise
3E-6 mW/cm2 at 75 Db sound level
9.6E-4 mW/cm2 at 100 Db sound level Direct Calculations from Acoustic TheoryPassive Human
Powered 1.8 mW (Shoe inserts >> 1 cm2) Published Study.
Thermal Conversion 0.0018 mW - 10 deg. C gradient Published Study.
Nuclear Reaction
80 mW/cm3
1E6 mWh/cm3 Published Data.
Fuel Cells
300 - 500 mW/cm3
~4000 mWh/cm3 Published Data.
Assume 1mW Average as definition of “Scavenged Energy”
7 Copyright 2004 Mani Srivastava
Will IC technology alone help?
Speed power efficiency has indeed gone up 10x / 2.5 years for Ps and DSPs in 1990s
degraded before 90s > 100 mW/MIP to < 1 mW/MIP since 1990
IC processes have provided 10x / 8 years since 1965 rest from power conscious IC design in recent years
Lower power for a given function & performance e.g. 1.6x / year reduction since early 80s for DSPs (source TI)
8 Copyright 2004 Mani Srivastava
But … Help from IC technology will slow down
e.g. circuit voltage reduction have provided big gains used to be 5V, now around 1.5-2V expected to plateau
Big gains from low-power IC design tricks behind us Strong indications of continued exponential increase in
operating frequency and # of functions we all want color displays, multimedia, wireless comm.,
speech recognition on our PDAs! increase of 10x / 7 yrs in gates, 10x/9 yrs in frequency
Power requirements of wireless communication functions also constrained by Shannon & Maxwell!
9 Copyright 2004 Mani Srivastava
IC Technology Trends and Power[source: De99 from Intel]
Transistor scaling Each generation of technology scaling results in 30% reduction in
minimum feature size now mostly “constant electric field” scaling
• gate length, effective electrical gate oxide thickness, and supply voltage scaled by 30%
• use to be constant voltage scaling previously Worst-case sub-threshold leakage current approx. constant
but, have been able to maintain approximately constant drive current per unit width of transistor, at reduced supply voltage
Net result: gate delay reduced by 30%
• areal component of junction capacitances reduced by 30% transistor density is doubled (I.e. die area reduced by 50%) total capacitance on chip reduced by 30% transistor capacitance per unit area increases by 43%
10 Copyright 2004 Mani Srivastava
IC Technology Trends and Power(contd.)
Power constant electric field scaling:
43% higher f, 30% less C, 30% less V energy per clock cycle reduces by 65%
• assumes unchanged ave. # of switching transitions per cycle power reduces by 50%
If can’t scale Vdd and Vt (e.g. if electric field < max for reliability) energy reduce by 30% power remains constant can’t add more transistors without increasing power budget! constant voltage scaling was employed until 0.8 micron
But, max power depends not only on technology, but also on implementation (size, frequency, circuit style, microarchitecture etc.)
Metric: “active” or “switched” capacitance (P/V2f) per unit area• scaling theory predicts 43% increase• real life: 30-35% (logic transistor density improvement < x2)
11 Copyright 2004 Mani Srivastava
IC Technology Trends and Power (contd.)
Clock frequency doubles every technology generation
instead of 43% (as predicted by 30% gate delay improvement) why?
ave # of gate delays in a clock cycle is reducing (more pipelined) advanced circuit techniques
Die size size of die increases 25% per technology generation new designs add more transistors (integration, complexity)
Interconnect scaling thinner, tighter higher resistance and capacitance more layers interconnect distribution (# of wires vs. length) does not change significantly
12 Copyright 2004 Mani Srivastava
Projections about Power Power dissipation of microprocessors
from 100W to 2,000W if supply voltage is scaled
supply current from 100A to 3000A
10,000W if supply voltage not scaled Die size will have to be restricted
15 mm die will mean power will stay around 100W and supply current around 300A
22 mm die will have 200W and 500A Leakage power ignored – but
increasingly important A key problem: technology scaling
still being done with delay as the primary goal (30% reduction) as opposed to a goal that incorporates power as well (e.g. Energy-Delay)
Source: Intel/[Flautner01]
1
10
100
1000
1.5 0.8 0.35 0.18 0.1
/Watts cm
2
Hot plate
Nuclear Reactor
13 Copyright 2004 Mani Srivastava
Barriers to Future Voltage Scaling
Voltage scaling requires threshold voltage Vt to be scaled as well (15% per generation) this increases sub-threshold leakage current impact on power consumption and circuit robustness
Leakage power total leakage current goes up 7.5x per generation leakage power power by 5x soon will become a significant portion of total
active power remains constant for constant die size leakage power, and therefore total power, can be substantially
reduced by cooling Essential to control die temperature power density (W/cm2): 0.6 micron chips surpassed a hot plate!
14 Copyright 2004 Mani Srivastava
Barriers to Future Voltage Scaling (contd.)
Circuit performance and robustness scaling V and Vt poses serious challenges to special
circuits such as domino logic, sense amplifiers increase in sub-threshold leakage current impacts bit line
delay on large on-chip caches divergence of logic and cache performance
Single event upsets: soft errors caused by alpha particles in material and cosmic rays reduced capacitance lower energy to flip a bit soft error rate will increase
• adding more capacitance undesirable• memory is typically ECC protected, but not latches, flip-flops
etc.
15 Copyright 2004 Mani Srivastava
Trends in Total Power Consumption
Frightening: proportional to area & frequency
DEC 21164
source : arpa-esto
microprocessorpower dissipation
16 Copyright 2004 Mani Srivastava
System Design for Low Power Need to explicitly design the system with power
consumption or energy efficiency in mind Fortunately, IC technology still continue to help
indirectly by increasing level of integration more and faster transistors can enable low-power
system architectures and design techniques• e.g. system integration on a chip can reduce the
significant circuit I/O power consumption
Energy efficient design of higher layers of the system also help
energy efficient protocols, power-aware apps. etc.
17 Copyright 2004 Mani Srivastava
System Design for Low Power (contd.)
Energy efficiency cuts across all system layers entire network, not just the node everything: circuit, logic, software, protocols, algorithms, user
interface, power supply... complex global optimization problem
Need to choose the right metric e.g. individual node vs. network lifetime
Trade-off between energy consumption & QoS optimize energy metric while meeting QoS constraint
Power-awareness, and not just low power right energy at the right time and place
19 Copyright 2004 Mani Srivastava
Power Supply
Where does the Power Go?B
atte
ry
DC-DCConverter
Communication
RadioModem
RFTransceiver
Processing
ProgrammablePs & DSPs
(apps, protocols etc.) Memory
ASICs
Peripherals
Disk Display
20 Copyright 2004 Mani Srivastava
Power Consumption for a Computer with Wireless NIC
Display36%
Wireless LAN18%
Hard Drive18%
CPU/Memory21%
Other7%
21 Copyright 2004 Mani Srivastava
Energy Consumption ofWireless NICs (Wavelan)
Specs Measured
2 Mbps(Bronze)
Sleep ModeIdle ModeReceive ModeTransmit Mode
9 mA--------280 mA330 mA
14 mA178 mA 200 mA280 mA
11 Mbps (Silver)
Sleep ModeIdle ModeReceive ModeTransmit Mode
10 mA--------180 mA280 mA
10 mA156 mA190 mA284 mA
22 Copyright 2004 Mani Srivastava
Power Consumption in Post-PC Devices
Pocket computers, PDAs, wireless pads, wireless sensors, pagers, cell phones
Energy and power usage of these devices is markedly different from laptop and notebook computers much wider dynamic range of power demand share of memory, communication and signal processing
subsystems become more important disk storage and displays disappear or become simpler
Design of power-aware higher layer applications and protocols need to be re-evaluated
23 Copyright 2004 Mani Srivastava
Example: Power Consumption for Berkeley’s InfoPad Terminal
DC/DC25%
LCD6%
I/O1%
Video Display
40%
Wireless18%
µProc.6%
Misc7%
With Optional Video DisplayTotal = 9.6W
(with processor at 7% duty cycle)
DC/DC42%
LCD10%
I/O2%
Wireless29%
µProc.6%
Misc11%
Without Optional Video DisplayTotal = 6.8W
(with processor at 7% duty cycle)
24 Copyright 2004 Mani Srivastava
Example: Power Consumption for Compaq WRL’s Itsy Computer
System power < 1W doing nothing (processor 95% idle)
107 mW @ 206 MHz 77 mW @ 59 MHz 62 mW @ 59 MHz, low voltage
MPEG-1 with audio 850 mW @ 206 MHz (16% idle)
Dictation 775 mW @ 206 MHz (< 0.5% idle)
text-to-speech 420 mW @ 206 MHz (53% idle) 365 mW @ 74 MHz, low voltage ( < 0.5% idle)
Processor: 200 mW 42-50% of typical total
LCD: 30-38 mW 15% of typical total
30-40% in notebooks
Itsy v1StrongARM 110059–206 MHz (300 us to switch)2 core voltages (1.5V, 1.23V)64M DRAM / 32M FLASHTouchscreen & 320x200 LCDcodec, microphone & speakerserial, IrDA
25 Copyright 2004 Mani Srivastava
Example: Power Consumption for Compaq’s iPAQ
206MHz StrongArm SA-1110 processor
320x240 resolution color TFT LCD
Touch screen
32MB SDRAM / 16MB Flash memory
USB/RS-232/IrDA connection
Speaker/Microphone
Lithium Polymer battery
PCMCIA card expansion pack & CF card expansion pack
* Note
CPU is idle state of most of its time
Audio, IrDA, RS232 power is measured when each part is idling
Etc includes CPU, flash memory, touch screen and all other devices
Frontlight brightness was 16
26 Copyright 2004 Mani Srivastava
The Power Hogs in Post-PC Devices
Wireless modem Rx is comparable, and may be more, than Tx
DC-DC conversion Displays
Handheld flat panel display6.4W Helmet mounted display 4.9W Integrated sight module display 2.6W
Digital subsystem for protocol & applications note: in InfoPad, the processor is used primarily for simple
management of transmit modem schedule Storage peripherals
27 Copyright 2004 Mani Srivastava
Energy Consumption in Wireless Sensor Nodes
Processing excluding low-level processing for radio, sensors, actuators
Radio Sensors Actuators Power supply
28 Copyright 2004 Mani Srivastava
Example: Sensor Nodes High-end sensor node: Rockwell
WINS nodes StrongARM 1100 processor Connexant’s RDSSS9M
900MHz DECT radio(100 kbps, ~ 100m)
Seismic sensor
Low-end sensor node: Berkeley’s COTS motes Atmel AS90LS8535
microcontroller RF Monolithic’s DR3000 radio
(2.4, 19.2, 115 kbps, ~ 10-30m)
light, temperature,10 kbps @ 20m
29 Copyright 2004 Mani Srivastava
Processing Common sensor node processors:
Atmel AVR, Intel 8051, StrongARM, XScale, ARM Thumb, SH Risc Power consumption all over the map, e.g.
16.5 mW for ATMega128L @ 4MHz 75 mW for ARM Thumb @ 40 MHz
But, don’t confuse low-power and energy-efficiency! Example
242 MIPS/W for ATMega128L @ 4MHz (4nJ/Instruction) 480 MIPS/W for ARM Thumb @ 40 MHz (2.1 nJ/Instruction)
Other examples: 0.2 nJ/Instruction for Cygnal C8051F300 @ 32KHz, 3.3V 0.35 nJ/Instruction for IBM 405LP @ 152 MHz, 1.0V 0.5 nJ/Instruction for Cygnal C8051F300 @ 25MHz, 3.3V 0.8 nJ/Instruction for TMS320VC5510 @ 200 MHz, 1.5V 1.1 nJ/Instruction for Xscale PXA250 @ 400 MHz, 1.3V 1.3 nJ/Instruction for IBM 405LP @ 380 MHz, 1.8V 1.9 nJ/Instruction for Xscale PXA250 @ 130 MHz, .85V (leakage!)
And, the above don’t even factor in operand size differences! However, need power management to actually exploit energy efficiency
Idle and sleep modes, variable voltage and frequency
30 Copyright 2004 Mani Srivastava
Radio Energy per bit in radios is a strong function of desired
communication performance and choice of modulation Range and BER for given channel condition (noise, multipath
and Doppler fading) Watch out: different people count energy differently
E.g. Mote’s RFM radio is only a transceiver, and a lot of low-level
processing takes place in the main CPU While, typical 802.11b radios do everything up to MAC and link
level encryption in the “radio” Transmit, receive, idle, and sleep modes Variable modulation, coding Currently around 150 nJ/bit for short rangeS
31 Copyright 2004 Mani Srivastava
Radio Power ConsumptionTx: Sender Rx: Receiver
ChannelIncominginformation
Outgoinginformation
TxelecE Rx
elecERFETransmit
electronicsReceive
electronicsPower
amplifier
0
2000
4000
6000
8000
0
100
200
300
0
200
400
600
TxelecE Rx
elecERFETxelecE Rx
elecERFETxelecE Rx
elecERFE
nJ/bit nJ/bit nJ/bit
~ 1 km (GSM) ~ 50 m (WLAN) ~ 10 m (Mote)
32 Copyright 2004 Mani Srivastava
Dominance of Electronics at Short Ranges
d
Static Power,Digital Processing
Power amp,Receiver Sensitivity
Radio Maximum
dn
2.4 KHz OOK(RFM TR1000 @ 916 MHz)
14 J 3.1 J
115.2 KHz ASK(RFM TR1000 @ 916 MHz)
372 nJ 65 nJ
1 Mbps Custom(MIT AMPS-1 @ 2.4 GHz)
570 nJ 740 nJ
11 Mbps 802.11b(Cisco Aironet 350 @ 2.4
GHz)236 nJ 91 nJ
54 Mbps 802.11a(Atheros, ISSCC2002)
14.8 nJ 11 nJ
€
Ebit =α + βdn
Re: Min et. al., Mobicom 2002 (Poster)
Sender Side Power Consumption
33 Copyright 2004 Mani Srivastava
Radio Electronics Trends
Analog electronics240 mW
Digital electronics170 mW
Power amplifier 600 mW
(~11% efficiency)
Intersil PRISM II (Nokia C021 wireless LAN)
Radiated power63 mW (18 dBm)
Trends: Move functionality from the analog to the digital electronics Digital electronics benefit most from technology improvements Analog a bottleneck Digital complexity still increasing (robustness)
34 Copyright 2004 Mani Srivastava
Computation & Communication Energy/bit Energy/op large even for short ranges!
Transmit 720 nJ/bit Processor 4 nJ/op
Receive 110 nJ/bit ~ 200 ops/bit
Mote-class Node
WINS-class Node Transmit 6600 nJ/bit Processor 1.6 nJ/op
Receive 3300 nJ/bit ~ 6000 ops/bit
TransmitReceive
Encode Decode Transmit
Receive
EncodeDecode
Energy breakdown for acoustic Energy breakdown for image
35 Copyright 2004 Mani Srivastava
Sensing Several energy consumption sources
transducer front-end processing and signal conditioning
analog, digital ADC conversion
Major source of power consumption Function of speed and resolution Fundamental limit on (speed*resolution/power) metric Important to select ADC for resolution that is truly required
Diversity of sensors: no general conclusions can be drawn Low-power modalities
Temperature, light, accelerometer Medium-power modalities
Acoustic, magnetic High-power modalities
Image, video, beamforming
36 Copyright 2004 Mani Srivastava
Actuation Emerging sensor platforms
Mounted on mobile robots Antennas or sensors that can be actuated
Energy trade-offs not yet studied Some thoughts:
Actuation often done with fuel, which has much higher energy density than batteries
E.g. anecdotal evidence that in some UAVs the flight time is longer than the up time of the wireless camera mounted on it
Actuation done during boot-up or once in a while may have significant payoffs
E.g. mechanically repositioning the antenna once may be better than paying higher communication energy cost for all subsequent packets
E.g. moving a few nodes may result in a more uniform distribution of node, and thus longer system lifetime
37 Copyright 2004 Mani Srivastava
Power Analysis of RSC’s WINS Nodes
Summary Processor
Active = 360 mW doing repeated
transmit/receive Sleep = 41 mW Off = 0.9 mW
Sensor = 23 mW Processor : Tx = 1 : 2 Processor : Rx = 1 : 1 Total Tx : Rx = 4 : 3 at
maximum range comparable at lower Tx
39 Copyright 2004 Mani Srivastava
Some Observations Using low-power components and trading-off unnecessary performance for power
savings can have orders of magnitude impact Node power consumption is strongly dependent on the operating mode
E.g. WINS consumes only 1/6-th the power when MCU is asleep as opposed to active
At short ranges, the Rx power consumption > T power consumption multihop relaying not necessarily desirable
Idle radio consumes almost as much power as radio in Rx mode Radio needs to be completely shut off to save power as in sensor networks idle
time dominates MAC protocols that do not “listen” a lot
Processor power fairly significant (30-50%) share of overall power In WINS node, radio consumes 33 mW in “sleep” vs. “removed”
Argues for module level power shutdown Sensor transducer power negligible
Use sensors to provide wakeup signal for processor and radio Not true for active sensors though…
40 Copyright 2004 Mani Srivastava
Metrics for Power
Power sets battery life in hours problem: power frequency (slow the system!)
Energy per operation fixes obvious problem with the power metric but can cheat by doing stuff that will slow the chip
– Energy/op = Power * Delay/op
Metric should capture both energy and performance: e.g. Energy/Op * Delay/Op
Energy*Delay = Power*(Delay/Op)2
41 Copyright 2004 Mani Srivastava
Communication vs. Computing A good measure is J/bit vs. J/ instruction Examples
Rockwell WINS nodes: 1500 to 2700 Medusa (similar to UCB’s motes): 220 to 2900 Sensoria’s WINS NG 2.0 nodes: ~ 1400
But watch out: not all instructions are the same: 8-bit vs. 32-bit not all bits are the same: distance, error probability
43 Copyright 2004 Mani Srivastava
Battery Characteristics
Important characteristics: energy density (Wh/liter) and specific energy (Wh/kg) power density (W/liter) and specific power (W/kg) open-circuit voltage, operating voltage cut-off voltage (at which considered discharged) shelf life (leakage) cycle life
The above are decided by “system chemistry” advances in materials and packaging have resulted in
significant changes in older systems– carbon-zinc, alkaline manganese, NiCd, lead-acid
new systems– primary and secondary (rechargeable) Li– secondary zinc-air, Ni-metal hydride
44 Copyright 2004 Mani Srivastava
Modeling the Battery Behavior
Theoretical capacity of battery is decided by the amount of the active material in the cell
batteries often modeled as buckets of constant energy – e.g. halving the power by halving the clock frequency is
assumed to double the computation time while maintaining constant computation per battery life
In reality, delivered or nominal capacity depends on how the battery is discharged
discharge rate (load current) discharge profile and duty cycle operating voltage and power level drained
45 Copyright 2004 Mani Srivastava
Battery Capacity
Current in “C” rating: load current nomralized to battery’s capacity e.g. a discharge current of 1C for a capacity of 500 mA-hrs is 500 mA
from [Powers95]
46 Copyright 2004 Mani Srivastava
Battery Capacity vs. Discharge Current
Amount of energy delivered is decreased as the current (rate at which power is drawn) is increased
rated as ampere hours or watt hours when discharged at a specific rate to a specific cut-off voltage
– primary cells rated at a current which is 1/100th of the capacity in ampere hours (C/100)
– secondary cells are rated at C/20 or C/10 At high currents, the diffusion process that moves new
active material from electrolytes to the electrode cannot keep up
concentration of active material at cathode drops to zero, and cell voltage goes down below cut-off
even though active material in cell is not exhausted!
47 Copyright 2004 Mani Srivastava
Battery Capacity vs. Discharge Current: Peukert’s Formula
Energy capacity: C = k/I k = constant dependent on chemistry & design = 0 for ideal battery (constant capacity), up to 0.7
for most loads in real batteries– also depends on chemistry and design
Good first order approximation does not capture effects of discharge profile
Battery life at constant voltage and currentL = C/P = C/(V.I) = (k/V).I-(1+)
49 Copyright 2004 Mani Srivastava
Amount of Computation during Battery Lifetime
Consider a system modification that changes performance by factor n and power by factor x
total work (= speed x lifetime) will change by n.x -(1+)
e.g. reducing the clock frequency by xN reduces power by xN (N>1) & reduces performance by xN,
work done changes by (1/N)x(1/N) -(1+) = N
– > 1 for >0 however, can’t just go on reducing frequency
– static power dissipated even at zero frequency– P = V.I = V.(S+Df)
optimum frequency to maximize computation problem: system performance does not change linearly with
frequency (e.g. memory bottlenecks)
50 Copyright 2004 Mani Srivastava
Alternate Equivalent View of the Battery
Manufacturer’s often give battery efficiency (%) vs. discharge rate (or discharge current ratio) discharge rate = Iave/Irated
Battery cannot respond to instantaneous changes in current so, a time constant used to calculate Iave
Given actual energy drawn by the circuit, one can use the battery efficiency to calculate the actual depletion in the stored energy in the battery
Example: battery efficiency is 60% and its rated capacity is 100 mAh @ 1V computed average DC-DC current of 300 mA would drain the
battery in 12 min, while at 100% efficiency it would last 1 hr
51 Copyright 2004 Mani Srivastava
Modeling Battery Efficiency
rated
aveI I
IR =
cyclebat TN
=
∑=
=batN
cyclesystem
batave cycleI
NI
0
)(1
cyclebatavebatbat TVIE )1( η−=
from [Simunic01]
52 Copyright 2004 Mani Srivastava
Digression:Metrics to Relate Power and Performance
MIPS/Watt: millions of instructions per Joule problem: running faster gives better MIPS/Watt increasing frequency by N
MIPS go up by xN power goes up < xN due to static power MIPS/Watt will increase!
W/Spec2 has similar problem Total computation during battery lifetime is better
shows diminishing returns of increasing frequency
53 Copyright 2004 Mani Srivastava
Capacity & Variable Discharge Current: Constant vs. Pulsed
Capacity can be extended by draining power in short discharge periods separated by rest periods
also works with constant background current Battery relaxes and partially recovers the active
material lost during the current impulse longer the rest period, the better is the recovery longer rest period needed as the discharge depth
becomes greater battery voltage also goes back up
54 Copyright 2004 Mani Srivastava
Benefits of Pulsed Discharge
Higher specific power for a given specific energy impulses of several times the limiting current value
can be obtained by choosing short pulses and long rest periods
Higher specific energy for a given specific power ideally, want specific energy = theoretical capacity depends on pulse and rest periods
55 Copyright 2004 Mani Srivastava
Exploiting Pulse Discharge
Gain in battery life if system shutdown is done taking into account the pulse discharge
Examples: protocols in case of radios where power during
transmission is a lot higher than during receive and idle periods
shutdown of CPUs and variable speed CPUs shutdown of disks
56 Copyright 2004 Mani SrivastavaSummary of BatteryNon-Idealities
Rate Capacity Effect Capacity depends on the discharge rate Higher discharge rate reduces the actual capacity
Relaxation Effect When discharge current is either cutoff or reduced, the battery’s
capacity is recovered
0 100 200 300 400 500 600
time (sec)
2.56
2.58
2.6
2.62
2.64
2.66
2.68
2.7
2.72
Voltage (V)
8.5 mA
1.9 mA
57 Copyright 2004 Mani Srivastava
Battery Modeling Predict battery lifetime given a load profile Many battery models
Model electrochemical processes in the battery Solve system of PDEs, e.g. Berkeley’s DUALFOIL Accurate but long simulation times and large number of
parameters (e.g. > 50 with DUALFOIL) Not easy to use in an optimization tool
Abstract representation of batteries E.g. Markov chain No physics or chemistry based justification Not easy to use in an optimization tool
Analytical models Capture key factors of battery performance E.g model by Rakhmatov & Vrudhula @ U. Arizona
59 Copyright 2004 Mani Srivastava
[Rakhmatov02] Battery Model:1-D Diffusion
Region of width w C(x,t) = concentration at t [0,L] at distance x [0,w] Concentration on electrode surface x=0 ? C* = initial concentration and let C(0,t) drops below the cutoff level Ccutoff at t=L Fick’s laws for concentration behavior due to 1-D diffusion
Where J(x,t) is the flux at time t at distance x, and D is the diffusion coefficient
60 Copyright 2004 Mani Srivastava
[Rakhmatov02] Battery Model:1-D Diffusion (contd.)
Boundary conditions Faraday’s laws: flux at electrode surface is proportional to
the current load i(t) Flux at the other boundary of the diffusion region x=w is
zero
Where A is the area of the electrode, v is the number of reacting electrons, and F is the Faraday’s constant
61 Copyright 2004 Mani Srivastava
[Rakhmatov02] Battery Model:1-D Diffusion (contd.)
There is an analytical solution to these pairs of PDEs and boundary conditions
Let:
In units of coulombs and 1/second respectively Then one gets the following battery model
62 Copyright 2004 Mani Srivastava
[Rakhmatov02] Battery Model:1-D Diffusion (contd.)
Two battery parameters: How to estimate them? Look at model again
RHS represents the capacity of the battery First term is the total charge consumed by the system Second term is the amount of charge in the battery that could
not be used by the system because it was not available at the electrode surface at the time of failure
As increases, the second term goes to zero Thus a large means that battery is practically an ideal source Intuitively, this is because a larger means a faster diffusion
63 Copyright 2004 Mani Srivastava
Estimating and
Approach: carry out a set of constant load tests For a constant current I, the model reduces to So, apply a set of constants loads I(1), … I(N) until
the battery is exhausted This results in a set of lifetime measurements
L(1), … L(N) Estimate and by minimizing the sum of
squares where
64 Copyright 2004 Mani Srivastava
Alternatives to Batteries?
Small batteries are the only choice for consumer products up to 20W
But heavy expensive expire without warning require replacement (disposal problem) or
recharging (time problem) Are there alternatives?
65 Copyright 2004 Mani Srivastava
No Batteries Needed!Energy Harvesting/Scavenging
Batteries are too big Batteries don’t last
forever Methods exist to extract
energy from the environment
Images from Smart Dust, Berkeley
Images from IASL, UWE, Bristol.
66 Copyright 2004 Mani Srivastava
Self-powered Chips Power generated using motion or solar cells, and
stored in a backup source (e.g. large capacitor) no batteries needed applicable to sensors on vehicles, body etc. e.g. Embedded power supply processor from MIT
[Amirtharajan97]
Back-up Source(large capacitor)
Generator
Processor
67 Copyright 2004 Mani SrivastavaExample: Scavenging from Motion
Media Lab’s “Parasitic Power Harvesting” project for devices built into a shoe http://www.media.mit.edu/resenv/power.html piezoelectric shoe inserts, shoe-mounted rotary
magnetic generator 20-80 mW of peak power during brisk walk, 1-2 mW average
a system had been built around the piezoelectric shoes that periodically broadcasts a 12-bit digital RFID as the bearer walks
charge stored in a capacitor over several footsteps and then discharged during RFID transmission every 3-5 footsteps
68 Copyright 2004 Mani Srivastava
Fuel Cells Invented in the 1990s: liberate energy from H atom Theoretically, quiet and clean like batteries Plus, amazing energetic potential
up to 20x more than NiCd of comparable size No length recharging: rapidly refueled Costs coming down considerably
sophisticated engineering, and reduced amount of expensive platinum required for catalysts
– while, $/J have gone up with energy-dense batteries example:
– NiCd weighs 0.5 kg, lasts 1 hr, and costs $20– comparable Li-Ion lasts 3 hrs, but costs > 4x more– comparable fuel cell will last 30 hrs, and may cost < $5
69 Copyright 2004 Mani Srivastava
Electrochemistry of Fuel Cells
ELECTROLYTE(specialized polymer
or other materialthat allows ions topass but blocks
electrons)
ANODE CATHODECATALYST(e.g. platinum)
HYDROGEN
OXYGEN
+
+
+
+
+
ELECTRONS
WATER
70 Copyright 2004 Mani Srivastava
Theoretical Energetic Potential of Fuel Cells
Stored Chemical Energy Wh/Kg Wh/liter
FUEL CELLSDecalin (C10H18) 2400 2100Liquid hydrogen 33000 2500Lithium borohydride (LiBH4 and4H20)
2800 2500
Solid metal hydride (LaNi5H6) 370 3300Methanol 6200 4900Hydrogen in graphite nanofibers 16,000 32,000
RECHARGEABLE BATTERIESLead acid 30 80NiCd 40 130Ni-metal hydride 60 200Lithium-ion 130 300
71 Copyright 2004 Mani Srivastava
Also Important: Modeling the DC-DC Converter Efficiency
The dependency of efficiency on the output current
cycleC
CC TV
EI =
DC
outbat
II
η=
cyclebatbatDCbat TVIE =
outDCbatDC EEE −=
from [Simunic01]
72 Copyright 2004 Mani Srivastava
Summary
Batteries improve very slowly The hunt for disruptive technology is on
Better sources Scavenging & harvesting
Until then: Exploit interesting non-linearities and dynamics that are
exhibited by batteries Resource allocation and scheduling can be made battery-
aware (task scheduling by OS in processors, packet scheduling by MAC in radios)
top related